An SEMI EDA Interface A implementation is more than a way to read values from a semiconductor tool. A Thai factory needs to specify which tool, module, event and sensor values it needs; the collection rate and trigger; the meaning and quality of each value; and who may change a collection plan. This guide turns those questions into a procurement specification and an acceptance test. It focuses on high-volume data acquisition, metadata, Data Collection Plans (DCPs), authorization and version compatibility, while preserving the control role of SECS/GEM.
The intended readers are fab procurement, process engineering and OT teams, as well as equipment suppliers asked to support Interface A. The SEMI standards do not change with the country of installation. What differs at a Thai site is the installed tool configuration, network boundary, operating team and support arrangement. Those details belong in the RFP.
Separate EDA data acquisition from SECS/GEM control
Two roles around the same tool
SEMI’s August 2026 explanation describes SECS/GEM and GEM300 mainly as control and configuration interfaces, while EDA/Interface A provides a parallel channel focused on high-volume, high-velocity data acquisition. An RFP should not instruct a supplier to replace SECS/GEM with EDA. Recipe operations, equipment state, jobs and material handling still need an agreed control interface. EDA should describe the analytical data to extract from the equipment. Our SECS/GEM implementation guide for Thailand covers that control interface in more depth.
Separate channels do not prove that collection has zero impact on the tool. Its CPU, storage, internal network and control software may share resources. A high-rate DCP must therefore be tested while the tool is operating in a representative production state. Ask the supplier for supported point counts, rate ranges, simultaneous clients, buffering and behavior when a consumer disconnects. The meaningful performance requirement is a combination of points, rates, latency, missing-data tolerance and tool load, not a single headline frequency.
SEMI notes that traditional GEM collection can be constrained because collection and control share a channel. EDA offers a more flexible route for event, trace and exception data. Actual performance still depends on the particular implementation and exposed data set. Do not infer support for every internal sensor from the words “EDA compatible.”
Define a read-only operating boundary accurately
We design EDA here as an analytical acquisition route: the analysis client must not issue recipe or production-control commands. That does not mean the acquisition interface has no management operations. A client or administrator must be able to create, activate, stop and maintain DCPs. The RFP should distinguish production-control commands from collection-plan administration and state who may perform each operation. Record approvals and audit logs for DCP changes.
If an analytics application is expected to stop a tool or alter a recipe after detecting a fault, that is a separate closed-loop control design. It requires additional safety analysis, change control and acceptance tests. Keep an initial acquisition PoC within its approved scope and review control feedback separately.
Specify the use case before listing standards
SEMI’s public Information & Control list includes E120, E125, E128, E132, E134, E138 and E164 among EDA-related standards. Broadly, E120 addresses the common equipment model, E125 equipment self-description, E132 client authentication and authorization, E134 data collection management, and E164 common metadata. Copying the numbers into a claim of “full compliance” is insufficient. Record the precise editions, implemented functions, equipment model and exposed data points. Use the official standards text and obtain a supplier conformance matrix for the purchased configuration.
Begin with two or three specific analytical tasks. Examples include correlating chamber pressure and valve states with a wafer or lot, viewing a temperature history around an alarm, and comparing transport events with downtime. “Collect everything for future AI” cannot set an upper bound on load, storage, retention or cost. For each task, name the tool, signal, required analysis and business decision.
Different tasks require different granularity. Monthly downtime classification may need events and state transitions rather than millisecond traces. Detecting a transient process excursion may require high-rate measurements and accurate timestamps. Comparing recipes needs recipe and step context. Build a use-case-by-signal collection matrix instead of imposing one sample interval everywhere.
Treat metadata as more than a tag list
Link tool structure to signal meaning
Equipment self-description is only useful when consumers can interpret it. Inspect how the installed tool represents equipment, modules, subsystems, sensors, events and parameters. Even tools with the same model number may differ by option and software edition. Ask for a metadata export from the actual configuration, not only a catalog screenshot.
For each point, request a stable identifier, display name, location in the equipment hierarchy, physical quantity, unit, data type, collection conditions, timestamp origin, missing-value representation, quality indicator, edition and change history. “Pressure” does not say which chamber it belongs to, whether it is absolute or gauge pressure, or whether its unit is Pa or Torr. “Temperature” may refer to a sensor reading, a setpoint or an estimate. These are data-dictionary requirements for the project; they are not a claim that a SEMI clause gives every supplier the same tag names.
Events, traces and exceptions also need production context. Check clock differences, module IDs, material IDs, recipes and process steps. Map which fields come from EDA, SECS/GEM, MES or another system. Do not assume EDA supplies every lot-history field. Our MQTT Sparkplug B factory integration guide discusses semantic continuity through data pipelines. Sparkplug B and EDA are distinct interfaces; a bridge between them needs explicit mapping and data-quality rules.
Store the accepted metadata as a baseline. After a tool software update or sensor replacement, compare IDs, units and hierarchy, identify affected DCPs and analytics models, and approve the change. Decide whether newly exposed points are collected automatically or only after review. A dashboard linked only by a mutable display name can silently show a different measurement after a rename or translation.
Manage the DCP lifecycle
A Data Collection Plan expresses a consumer’s collection request to equipment. Public SEMI-related material discusses event and trace requests and conditional collection. In the RFP, require a demonstration of the actual implementation’s create, validate, activate, stop, restart and delete behavior. “Supports DCP” is a starting statement, not acceptance evidence. The audit trail should show who changed a plan, when and why.
Keep plans aligned with uses such as availability monitoring, process control diagnostics and maintenance. One huge plan obscures ownership and load. For each plan, specify its owner, points, trigger events, trace intervals, start and stop conditions, priority and change request. Document the destination and retention period alongside that DCP as requirements for the factory collector and data pipeline; do not present them as mandatory native E134 DCP settings. Test gaps and overlaps when plans switch. Where several clients request the same signal, verify the tool’s limits and load behavior.
Point count alone cannot size a solution. Five hundred small values once per second differ from five hundred structured values at high frequency. Estimate data volume from point count, average size, frequency, event bursts and protocol overhead, then measure it in the PoC. Test a tool stop and restart, client disconnect and full buffer. Document which data are lost, duplicated or delayed. Do not turn an illustrative calculation into a supplier performance guarantee.
Combine authorization with the OT network design
E132 addresses client authentication and authorization. It is not a license to expose a data port to any consumer. Define client identities, approved sources, certificates or other chosen authentication, roles, credential renewal and revocation, and audit records. Determine whether collection read access and DCP modification can be separated. Route changes to production collection through the same quality and equipment change process used by the factory.
Draw the tool server, collector, OT DMZ, storage and analytics destination on one network diagram, with traffic directions. A proposal to connect the tool directly to a cloud service needs review of factory security rules, contractual data rights and storage location. It may be preferable to collect inside the plant and restrict what leaves it. Our industrial data diode selection guide describes higher-security one-way boundaries; a one-way device alone does not solve how DCP administration would work across that boundary.
Before FAT, assign owners for time synchronization, DNS, firewall rules, certificate expiry and logs. At a Thai site, the equipment vendor, local integrator, site OT team and head-office IT group may each control one piece. An RACI table should say who renews a certificate and who responds to a night-shift outage. Otherwise the initial project can work while collection fails at the first credential renewal.
Verify editions instead of buying a “Freeze 3” label
SEMI’s 2018 article described the existing SOAP/XML over HTTP/1.1 approach and proposed HTTP/2, gRPC and Protocol Buffers. Its 2020 article reported publication of E179-0320 and continuing work on other EDA standards as part of proposed Freeze 3 development. These are dated accounts of standards work. They do not, by themselves, prove that the whole Freeze 3 suite was ratified and interoperable in 2026. Check current official SEMI editions and each supplier’s implemented feature set at procurement time.
Make a compatibility matrix with standard number and edition, transport, encoding, authentication, metadata retrieval, DCP operations, event/trace/exception data, simultaneous clients and error handling. Put the tool server, factory collector and analytics adapter in separate columns. Run interoperability tests on the exact versions being purchased. If a converter is necessary, include its development, maintenance and regression testing in the quotation.
Standards conformance does not disclose all model-specific points. Ask the equipment supplier for the actual metadata export, option differences, maximum tested collection load, software-update policy and anonymized interoperability evidence. If a customer requirement names a particular edition or Freeze, quote it accurately rather than replacing it with a vague promise to support the “latest.”
Make the PoC prove useful data, not just connectivity
Freeze the test tool, software version, collector version, points, DCP, network and work window before beginning. For a production tool, obtain owner approval for an impact assessment and rollback procedure. Acceptance should cover metadata completeness, units, event and trace timestamps, lot and process context, peak missing data, DCP audit records, tool-control impact and disconnect recovery. A live value on a screen proves very little.
Measure expected and observed values by point, collection-interval distribution, end-to-end latency distribution, missing, duplicate and retransmitted records, clock offset, tool CPU, memory and network load, and any quality impact. Averages can hide a short burst or a gap at a process transition. Test recipe changes, alarms, maintenance, tool stops and client restarts. Save raw examples and compare record counts, sequence and units programmatically as well as visually.
Set numerical acceptance thresholds from the use case. In-process anomaly detection may have very different latency needs from daily improvement reporting. There is no universal latency or loss number to insert into every EDA contract. Ask the equipment supplier for its supported range and compare it with the factory’s required range. If the PoC fails, price alternatives such as fewer points, conditional DCPs, additional tool resources or a phased analytical use case.
Write an RFP suppliers can answer consistently
Start with the decision the data must support. Then specify tool and options, existing SECS/GEM connection, points, editions and conformance, metadata, DCP, authorization, network, performance, storage, tests and support. Use a responsibility table for the tool, collection client and MES/FDC analytics. Give each bidder the same inputs.
Separate quotation lines for tool EDA licenses, software upgrades, point exposure, metadata preparation, collector, network and credentials, storage, integration, validation, training and maintenance. Ask for first-tool and subsequent-tool pricing separately. A tool may support EDA but charge for exposing additional points. The service agreement should cover point and unit changes, compatibility testing after updates, diagnostic logs and local or remote response times.
Require each answer to say “standard,” “configuration,” “custom development” or “unsupported,” and to cite product edition, manual section or demonstration evidence. A free-text “possible” is hard to compare. Include the vendor’s referenced standard editions and the person responsible for its conformance statement.
| RFP item | Supplier response | Acceptance evidence |
|---|---|---|
| Editions and compatibility | SEMI edition, transport, encoding, connection constraints | Conformance matrix and logs |
| Metadata | Stable ID, hierarchy, type, unit and change notification | Live export and change diff |
| DCP | Lifecycle, roles and concurrent-plan limits | Operation log and load results |
| Data quality | Rate, latency, loss, clock and duplicates | Production scenario measurement |
| Security | Authentication, role separation and credential renewal | Revocation and denial tests |
| Recovery | Tool/client restart, network loss and buffer behavior | Recovery record and data comparison |
Use FAT and SAT for different evidence
At FAT, use the agreed tool and collector versions in the supplier’s factory or test environment. Exercise metadata, DCPs, authorization, event and trace data, load and disconnect recovery. Save logs and raw data. If full production conditions cannot be reproduced, list simulated conditions and open tests. Agree which FAT cases must be rerun after a software change.
At SAT in Thailand, use the installed OT network, real credentials, time service, MES/FDC connections and production state. A design that passed FAT may fail because of firewall rules, DNS, network bandwidth, time drift or option differences. Test product changeover, alarms, maintenance, tool stop and a controlled communications outage. If an operation cannot be interrupted, agree on a safe alternative or staged acceptance and record the untested case as open.
For every requirement ID, record method, expected result, observation, evidence file, pass/fail, issue owner and due date. Screenshots alone are weak evidence; attach timestamped logs, raw records and tool load measurements. When values go missing or change meaning, locate the cause in the tool, network, collector or analytics transformation before accepting a workaround. Handover must include production monitoring and escalation contacts.
Sustain the installation at a Thai site
Even if head office funds the project, local maintenance and process engineers will operate it. Provide the point dictionary, DCP list and recovery procedures in an operating language they can use. Define the responsibilities of the tool supplier, collector vendor, local integrator, site OT and process team. Assign both a technical owner for each point and a business owner for its analytical use.
Retrofit work requires further checks: whether the installed tool implements the required EDA edition, whether CPU or storage upgrades are needed, whether warranty is affected, and how much outage is available. Prioritize tools by data value, integration effort and allowable downtime. Use the first installation’s measurements to revise the rollout standard before updating a fleet.
After go-live, review unused DCPs, missing-data rates, latency, metadata changes, authentication failures and storage growth. When a process improvement relies on a data set, retain the tool version and dictionary version that produced it. Include periodic review and change procedures in the initial contract.
Tie version migration to an outage plan
A supplier’s offer to upgrade EDA is not by itself a reason to update a running fleet. Identify the operational gain first: a missing point becomes available, measured traffic drops, authentication meets factory policy, or support can be extended. Then identify changes to the tool server, collector, adapter, DCPs, certificates and monitoring. If the gain is modest but the upgrade touches several active lines, begin with one bounded installation.
Before migration, back up accepted metadata, DCPs, configuration and sample records. Compare identifiers and units after the update, test the exact client pair, and define a rollback trigger. Collect comparable data from the same product and process conditions before and after the change to detect gaps or changed meanings. If rollback is not supported, agree on recovery steps and maximum outage before signing. A newer edition should be judged against its data value, production risk and maintenance cost.
Record concrete acceptance failures
A hierarchy may pass FAT, then a tool with a different option at SAT may expose changed sensor IDs. Require an explanation of the mapping and future ID stability instead of silently relabeling the data. A pressure point shown in Pa at FAT and Torr at SAT cannot go directly to analysis. If conversion is needed, retain the raw value and document where conversion occurs. A visible zero must be distinguishable from a true zero, sensor failure and missing communication.
Normal high-rate traces can look sound while the brief period before an alarm is missing, making fault diagnosis unreliable. Align DCP switch times, tool load, retransmission and collector write queues to locate the gap. Preserve equipment and receipt timestamps to investigate out-of-order records. When data are replayed after a network break, define deduplication keys and allowable lateness. Accept on raw loss, duplicate and latency evidence, not on a smooth-looking chart.
Security acceptance should test denial as well as successful login: revoked credentials, an unapproved terminal and a read-only operator trying to change a DCP. Test reconnect after certificate rotation, DCP state after tool restart and audit-log preservation. If malformed requests stop collection or affect production control, investigate load and exception handling as well as permissions. Each failure needs reproduction steps, impact, workaround, permanent correction and retest deadline, with an explicit decision on production start.
Frequently asked questions
Does SEMI EDA Interface A replace SECS/GEM?
No. SEMI describes them as parallel roles: SECS/GEM and GEM300 mainly support equipment control and configuration; EDA mainly supports high-volume acquisition. Specify the boundary for the actual tool and factory host.
Can an EDA-compatible tool expose every internal sensor?
Not necessarily. Verify actual points, units, metadata, collection conditions, options and software edition. A sensor’s physical presence does not guarantee its external publication.
Does a DCP permit unlimited high-rate collection?
No. Point count, rate, data type, client count, tool resources, network and storage impose limits. Test the intended peak, missing records and control impact with the vendor.
Should the RFP simply demand Freeze 3?
Identify the customer’s requirements and exact tool/client editions. Public development articles alone cannot prove that every related edition is ratified and interoperable. Specify the formal editions and features, then test the pair.
Where do costs differ most?
Compare the included tool license, point development, upgrades, collector, security and network work, storage, MES/FDC integration, FAT/SAT and maintenance. Quotes are meaningful only when they cover the same scope.
Conclusion
The goal of an SEMI EDA Interface A project is to acquire meaningful, timely process data without disrupting production. Define the control/acquisition boundary, require a usable metadata dictionary and manageable DCPs, check authorization and version compatibility, and measure both quality and tool load in a PoC. Follow the same requirement IDs through FAT and site acceptance in Thailand.
If you are preparing a tool supplier RFP, organizing points on an installed tool, or defining PoC and FAT/SAT evidence, contact TOMAS TECH. We can start from the intended data and decision even before a particular tool or product has been selected.
Sources
- SEMI Information & Control standards list
- SEMI: Next-Generation EDA Standards
- SEMI: More Data, Faster, with the Next Generation of EDA Standards
- SEMI: Newly Formed Diagnostic Data Acquisition Task Force in China (2026)
- SEMI: Everything You Need to Know about the EDA Standards Suite
- SEMI: E30, GEM and EDA
